Housekeeping

Introduction

The purpose of this exercise is to give you additional practice working with logistic regression. The exercise is based on the NBA Team Game Stats dataset found here: https://www.kaggle.com/ionaskel/nba-games-stats-from-2014-to-2018/. Read more about the problem and dataset under the description section of the link.

Kaggle is a great online community of data scientists. To learn more about Kaggle, follow this link: https://www.kaggle.com/getting-started/44916.

For this exercise, you will work with the same NBA team as last time. As before, you will need to set aside data for the 2017/2018 season as test data. Write the name of the team you selected at the top of your report.

The data

This is the same data from the last in-class exercise. so you should already have it saved locally. Just in case you do not, follow the instructions below.

Download the data (named nba_games_stats.csv) from Sakai and save it locally to the same directory as your R markdown file. To find the data file on Sakai, go to Resources \(\rightarrow\) Datasets \(\rightarrow\) In-Class Analyses. Once you have downloaded the data file into the SAME folder as your R markdown file, load and clean the data by using the following R code.

⊕Again, you are expected to select ONLY ONE NBA team in the data. Also, set aside data for the 2017/2018 season as test data.

nba <- read.csv("nba_games_stats.csv",header = TRUE,sep = ",",stringsAsFactors = FALSE)

# Set factor variables
nba$Home <- factor(nba$Home)
nba$Team <- factor(nba$Team)
nba$WINorLOSS <- factor(nba$WINorLOSS)

# Convert date to the right format
nba$Date <- as.Date(nba$Date, "%Y-%m-%d")

# Also create a binary variable from WINorLOSS. 
# This is not always necessary but can be useful 
#particularly for R functions that prefer numeric binary variables
#to the original factor variables
nba$Win <- rep(0,nrow(nba))
nba$Win[nba$WINorLOSS=="W"] <- 1

# I picked the Charlotte Hornets (CHO) as an example,
#you should pick any team you want
nba_reduced <- nba[nba$Team == "CHO", ]

# Set aside the 2017/2018 season as your test data
nba_reduced_train <- nba_reduced[nba_reduced$Date < "2017-10-01",]
nba_reduced_test <- nba_reduced[nba_reduced$Date >= "2017-10-01",]

You will now use both the nba_reduced_train and nba_reduced_test datasets in this in-class exercise.

Exercises

Once again, treat the variable Win (or WinorLoss) as your response variable and the other variables as potential predictors. First fit the same model from the last in-class exercise using the nba_reduced_train data.

1. Add Opp.FieldGoals. as a predictor to the previous model from last time. Is the coefficient significant? If yes, interpret the coefficient in the context of the question.

2. What is the accuracy of this new model? Plot the roc curve for the fitted model. What is the new AUC value? Which model predicts the odds of winning better? The model from Question 1 or the model from the last in-class exercise?

3. Using the results of the model with the better predictive ability, what suggestions do you have for the coach of your team trying to improve the odds of his team winning a regular season game?

4. Use this “better” model to predict out-of-sample probabilities for the nba_reduced_test data. Using 0.5 as your cutoff for predicting wins or losses (1 vs 0) from the out-of-sample predicted probabilities, what is the out-of-sample accuracy? How well does your model do in predicting data for the 2017/2018 season?

5. Using the change in deviance test (anova test), test whether including Opp.Assists and Opp.Blocks in this “better” model at the same time would improve the model. Is there any other variable in this dataset which we did not consider that you think might improve our model? Which one and why?

Acknowledgement

This exercise is based on ideas proposed by Sam Voisin.